dgenerate-ultralytics-headless 8.3.239__py3-none-any.whl → 8.3.241__py3-none-any.whl

{dgenerate_ultralytics_headless-8.3.239.dist-info → dgenerate_ultralytics_headless-8.3.241.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dgenerate-ultralytics-headless
-Version: 8.3.239
+Version: 8.3.241
 Summary: Automatically built Ultralytics package with python-opencv-headless dependency instead of python-opencv
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>
@@ -51,12 +51,13 @@ Requires-Dist: pytest; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
 Requires-Dist: coverage[toml]; extra == "dev"
 Requires-Dist: zensical>=0.0.9; python_version >= "3.10" and extra == "dev"
-Requires-Dist: mkdocs-ultralytics-plugin>=0.2.3; extra == "dev"
+Requires-Dist: mkdocs-ultralytics-plugin>=0.2.4; extra == "dev"
 Requires-Dist: minijinja>=2.0.0; extra == "dev"
 Provides-Extra: export
 Requires-Dist: numpy<2.0.0; extra == "export"
 Requires-Dist: onnx>=1.12.0; platform_system != "Darwin" and extra == "export"
 Requires-Dist: onnx<1.18.0,>=1.12.0; platform_system == "Darwin" and extra == "export"
+Requires-Dist: onnxslim>=0.1.80; extra == "export"
 Requires-Dist: coremltools>=9.0; (platform_system != "Windows" and python_version <= "3.13") and extra == "export"
 Requires-Dist: scikit-learn>=1.3.2; (platform_system != "Windows" and python_version <= "3.13") and extra == "export"
 Requires-Dist: openvino>=2024.0.0; extra == "export"

{dgenerate_ultralytics_headless-8.3.239.dist-info → dgenerate_ultralytics_headless-8.3.241.dist-info}/RECORD

@@ -1,4 +1,4 @@
-dgenerate_ultralytics_headless-8.3.239.dist-info/licenses/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
+dgenerate_ultralytics_headless-8.3.241.dist-info/licenses/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
 tests/__init__.py,sha256=bCox_hLdGRFYGLb2kd722VdNP2zEXNYNuLLYtqZSrbw,804
 tests/conftest.py,sha256=mOy9lGpNp7lk1hHl6_pVE0f9cU-72gnkoSm4TO-CNZU,2318
 tests/test_cli.py,sha256=GhIFHi-_WIJpDgoGNRi0DnjbfwP1wHbklBMnkCM-P_4,5464
@@ -8,7 +8,7 @@ tests/test_exports.py,sha256=8dxll33sqhlHeWplly5xxSdT-Nqw2cRRZWN3jZeRWWc,14155
 tests/test_integrations.py,sha256=6QgSh9n0J04RdUYz08VeVOnKmf4S5MDEQ0chzS7jo_c,6220
 tests/test_python.py,sha256=viMvRajIbDZdm64hRRg9i8qZ1sU9frwB69e56mxwEXk,29266
 tests/test_solutions.py,sha256=CIaphpmOXgz9AE9xcm1RWODKrwGfZLCc84IggGXArNM,14122
-ultralytics/__init__.py,sha256=Kkc0zFGRthNdbe2UViHpuowgEH5NVweklAPQWaMxVHg,1302
+ultralytics/__init__.py,sha256=S3XtXs_EbjQDpks7nIo7kSOhs1K6f8di-vl6nLmJ6_g,1302
 ultralytics/py.typed,sha256=la67KBlbjXN-_-DfGNcdOcjYumVpKG_Tkw-8n5dnGB4,8
 ultralytics/assets/bus.jpg,sha256=wCAZxJecGR63Od3ZRERe9Aja1Weayrb9Ug751DS_vGM,137419
 ultralytics/assets/zidane.jpg,sha256=Ftc4aeMmen1O0A3o6GCDO9FlfBslLpTAw0gnetx7bts,50427
@@ -123,7 +123,7 @@ ultralytics/data/scripts/get_coco.sh,sha256=UuJpJeo3qQpTHVINeOpmP0NYmg8PhEFE3A8J
 ultralytics/data/scripts/get_coco128.sh,sha256=qmRQl_hOKrsdHrTrnyQuFIH01oDz3lfaz138OgGfLt8,650
 ultralytics/data/scripts/get_imagenet.sh,sha256=hr42H16bM47iT27rgS7MpEo-GeOZAYUQXgr0B2cwn48,1705
 ultralytics/engine/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6DXppv1-QUM,70
-ultralytics/engine/exporter.py,sha256=IRUN6CNxBV9Gw8Qy-MZb_5p5bT2-njF91LVywa6FOss,72932
+ultralytics/engine/exporter.py,sha256=XG7JOdeJb_4pQF-iQkf1pCTCsFn_VVLILhBI_wu-iPM,73038
 ultralytics/engine/model.py,sha256=RkjMWXkyGmYjmMYIG8mPX8Cf1cJvn0ccOsXt03g7tIk,52999
 ultralytics/engine/predictor.py,sha256=neYmNDX27Vv3ggk9xqaKlH6XzB2vlFIghU5o7ZC0zFo,22838
 ultralytics/engine/results.py,sha256=LHX0AaVOv3CEjYjw8i4LThXqihxmahWCxpH20b4s9dM,68030
@@ -155,7 +155,7 @@ ultralytics/models/sam/amg.py,sha256=aYvJ7jQMkTR3X9KV7SHi3qP3yNchQggWNUurTRZwxQg
 ultralytics/models/sam/build.py,sha256=rEaFXA4R1nyutSonIenRKcuNtO1FgEojnkcayo0FTP4,12867
 ultralytics/models/sam/build_sam3.py,sha256=Gg_LiqNrCDTYaDWrob05vj-ln2AhkfMa5KkKhyk5wdE,11976
 ultralytics/models/sam/model.py,sha256=cOawDSkFqJPbt3455aTZ8tjaoWshFWFHQGGqxzsL_QQ,7372
-ultralytics/models/sam/predict.py,sha256=g-p88aNSeWql6bsjRb5FmjOXHGKvG0aJ9aIIlub7juU,203078
+ultralytics/models/sam/predict.py,sha256=rvUXBzNMvKlfjWP6UoKNm9GGpLcxJGsZGTkFq-7UOhI,203675
 ultralytics/models/sam/modules/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6DXppv1-QUM,70
 ultralytics/models/sam/modules/blocks.py,sha256=ZU2aY4h6fmosj5pZ5EOEuO1O8Cl8UYeH11eOxkqCt8M,44570
 ultralytics/models/sam/modules/decoders.py,sha256=G4li37ahUe5rTTNTKibWMsAoz6G3R18rI8OPvfunVX8,25045
@@ -164,7 +164,7 @@ ultralytics/models/sam/modules/memory_attention.py,sha256=jFVWVbgDS7VXPqOL1e3gAz
 ultralytics/models/sam/modules/sam.py,sha256=-KV-1PZK39DTdSpR5DI3E8I6gGVLja3tMv1MH7Au_eA,61654
 ultralytics/models/sam/modules/tiny_encoder.py,sha256=RJQTHjfUe2N3cm1EZHXObJlKqVn10EnYJFla1mnWU_8,42065
 ultralytics/models/sam/modules/transformer.py,sha256=NmTuyxS9PNsg66tKY9_Q2af4I09VW5s8IbfswyTT3ao,14892
-ultralytics/models/sam/modules/utils.py,sha256=ztihxg0ssx0W-CKiqV-8KzB4og39TKnbmV3YO96ENPw,20770
+ultralytics/models/sam/modules/utils.py,sha256=Re09hcKe9LJpFzKHxpbwEmoMpPImnJaoNw7pe62_ui0,21129
 ultralytics/models/sam/sam3/__init__.py,sha256=aM4-KimnYgIFe-e5ctLT8e6k9PagvuvKFaHaagDZM7E,144
 ultralytics/models/sam/sam3/decoder.py,sha256=kXgPOjOh63ttJPFwMF90arK9AKZwPmhxOiexnPijiTE,22872
 ultralytics/models/sam/sam3/encoder.py,sha256=zP934Q_7B2I9UJvQY4Ghimu7Y8dQjBCGqUmSv5zbsq0,21794
@@ -210,7 +210,7 @@ ultralytics/models/yolo/yoloe/train.py,sha256=giX6zDu5Z3z48PCaBHzu7v9NH3BrpUaGAY
 ultralytics/models/yolo/yoloe/train_seg.py,sha256=0hRByMXsEJA-J2B1wXDMVhiW9f9MOTj3LlrGTibN6Ww,4919
 ultralytics/models/yolo/yoloe/val.py,sha256=utUFWeFKRFWZrPr1y3A8ztbTwdoWMYqzlwBN7CQ0tCA,9418
 ultralytics/nn/__init__.py,sha256=538LZPUKKvc3JCMgiQ4VLGqRN2ZAaVLFcQbeNNHFkEA,545
-ultralytics/nn/autobackend.py,sha256=nenk1TkSzn-toSM223ukEdgV45d4IaSdolmvTlHs8r0,44410
+ultralytics/nn/autobackend.py,sha256=mCPRyW3OGt_jzi1dwAJXH93ntsHqg_95AfSjWhjt-cw,44612
 ultralytics/nn/tasks.py,sha256=636MN27VvlupIaHPpV4r0J705RUdV-giNKjIeJbRkxI,70448
 ultralytics/nn/text_model.py,sha256=novnuosqXnW1NmlOzWOk7dEKuN6Vq40CTksr6hI3Knc,15109
 ultralytics/nn/modules/__init__.py,sha256=5Sg_28MDfKwdu14Ty_WCaiIXZyjBSQ-xCNCwnoz_w-w,3198
@@ -250,7 +250,7 @@ ultralytics/trackers/utils/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6D
 ultralytics/trackers/utils/gmc.py,sha256=cvvhNXOhylVQti4pJQSNPx4yPqhhhw1k2yzY0JFl7Zo,13760
 ultralytics/trackers/utils/kalman_filter.py,sha256=crgysL2bo0v1eTljOlP2YqIJDLBcHjl75MRpbxfaR_M,21514
 ultralytics/trackers/utils/matching.py,sha256=jRZQrYSECQEwILn3mr7OodB8o7GawV-1v869plHstwk,7147
-ultralytics/utils/__init__.py,sha256=Fxq65P02ZEKQF1I4SEtU9MTVv-__fBCYg7JSl9FgNbk,55126
+ultralytics/utils/__init__.py,sha256=JfvODTB4mG_JOhTeCiPtq0iCEgiCh14hJf195rnOhLQ,55145
 ultralytics/utils/autobatch.py,sha256=jiE4m_--H9UkXFDm_FqzcZk_hSTCGpS72XdVEKgZwAo,5114
 ultralytics/utils/autodevice.py,sha256=rXlPuo-iX-vZ4BabmMGEGh9Uxpau4R7Zlt1KCo9Xfyc,8892
 ultralytics/utils/benchmarks.py,sha256=S_W4S4pe2ktSRdSuWb6m09UEFQmZhmjl943bbo67hOI,32277
@@ -269,7 +269,7 @@ ultralytics/utils/metrics.py,sha256=pP1LoKweL3Iga5dvCBVoqMSi5TwaaHERKKugDUloUs8,
 ultralytics/utils/nms.py,sha256=zv1rOzMF6WU8Kdk41VzNf1H1EMt_vZHcbDFbg3mnN2o,14248
 ultralytics/utils/ops.py,sha256=Se_Vb_smotVZ4g9gW4x56LpflAmbNDCEaeQz5BmUiD0,25637
 ultralytics/utils/patches.py,sha256=mD3slAMAhcezzP42_fOWmacNMU6zXB68Br4_EBCyIjs,7117
-ultralytics/utils/plotting.py,sha256=ivlhx4PZP80O-E8r4ruwxLYQEc6BLq5fmveKHgSi-EA,48219
+ultralytics/utils/plotting.py,sha256=nLProk2U3xYBwV4CdkeBx1mrPnrPA550cNSQRGbcLsg,48368
 ultralytics/utils/tal.py,sha256=w7oi6fp0NmL6hHh-yvCCX1cBuuB4JuX7w1wiR4_SMZs,20678
 ultralytics/utils/torch_utils.py,sha256=zOPUQlorTiEPSkqlSEPyaQhpmzmgOIKF7f3xJb0UjdQ,40268
 ultralytics/utils/tqdm.py,sha256=5PtGvRE9Xq8qugWqBSvZefAoFOnv3S0snETo5Z_ohNE,16185
@@ -291,8 +291,8 @@ ultralytics/utils/export/__init__.py,sha256=Cfh-PwVfTF_lwPp-Ss4wiX4z8Sm1XRPklsqd
 ultralytics/utils/export/engine.py,sha256=23-lC6dNsmz5vprSJzaN7UGNXrFlVedNcqhlOH_IXes,9956
 ultralytics/utils/export/imx.py,sha256=UHIq_PObOphIxctgSi0-5WaHvolHsHd3r5TTSjQSdgo,12860
 ultralytics/utils/export/tensorflow.py,sha256=igYzwbdblb9YgfV4Jgl5lMvynuVRcF51dAzI7j-BBI0,9966
-dgenerate_ultralytics_headless-8.3.239.dist-info/METADATA,sha256=PgCv3h3IXNlDfEVMUUSY2zWtTEV0HThO2cyZfavlRak,38747
-dgenerate_ultralytics_headless-8.3.239.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-dgenerate_ultralytics_headless-8.3.239.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
-dgenerate_ultralytics_headless-8.3.239.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
-dgenerate_ultralytics_headless-8.3.239.dist-info/RECORD,,
+dgenerate_ultralytics_headless-8.3.241.dist-info/METADATA,sha256=9qvOTFpuIuFvwutj2DJnpJkcuGoXmoIulGzzZ-e6gw8,38798
+dgenerate_ultralytics_headless-8.3.241.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+dgenerate_ultralytics_headless-8.3.241.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
+dgenerate_ultralytics_headless-8.3.241.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
+dgenerate_ultralytics_headless-8.3.241.dist-info/RECORD,,

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.239"
+__version__ = "8.3.241"
 
 import importlib
 import os

ultralytics/engine/exporter.py

@@ -1205,8 +1205,9 @@ class Exporter:
         # Setuptools bug: https://github.com/pypa/setuptools/issues/4483
         check_requirements("setuptools<71.0.0")  # Setuptools bug: https://github.com/pypa/setuptools/issues/4483
         check_requirements(("executorch==1.0.1", "flatbuffers"))
+        # Pin numpy to avoid coremltools errors with numpy>=2.4.0, must be separate
+        check_requirements("numpy<=2.3.5")
 
-        import torch
         from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
         from executorch.exir import to_edge_transform_and_lower
 

ultralytics/models/sam/modules/utils.py

@@ -210,9 +210,14 @@ def apply_rotary_enc(
         # No keys to rotate, due to dropout
         return xq_out.type_as(xq).to(xq.device), xk
     # Repeat freqs along seq_len dim to match k seq_len
-    if repeat_freqs_k:
-        r = xk_.shape[-2] // xq_.shape[-2]
-        freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 2)), r, 1)
+    if repeat_freqs_k and (r := xk_.shape[-2] // xq_.shape[-2]) > 1:
+        # MPS doesn't support repeat on complex tensors, decompose to real representation
+        if freqs_cis.device.type == "mps":
+            freqs_cis = torch.view_as_real(freqs_cis)
+            freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 3)), r, 1, 1)
+            freqs_cis = torch.view_as_complex(freqs_cis.contiguous())
+        else:
+            freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 2)), r, 1)
     xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)
     return xq_out.type_as(xq).to(xq.device), xk_out.type_as(xk).to(xk.device)
 

ultralytics/models/sam/predict.py

@@ -878,6 +878,7 @@ class SAM2VideoPredictor(SAM2Predictor):
         self.clear_non_cond_mem_around_input = False
         self.clear_non_cond_mem_for_multi_obj = False
         self.callbacks["on_predict_start"].append(self.init_state)
+        self.clear_non_cond_mem = True  # Whether to clear non-conditioning memory periodically
 
     def get_model(self):
         """Retrieve and configure the model with binarization enabled.
@@ -952,6 +953,7 @@ class SAM2VideoPredictor(SAM2Predictor):
                 run_mem_encoder=True,
             )
             output_dict[storage_key][frame] = current_out
+            self._prune_non_cond_memory(frame)
         # Create slices of per-object outputs for subsequent interaction with each
         # individual object after tracking.
         self._add_output_per_object(frame, current_out, storage_key)
@@ -1830,6 +1832,25 @@ class SAM2VideoPredictor(SAM2Predictor):
         inference_state["frames_already_tracked"].clear()
         inference_state["first_ann_frame_idx"] = None
 
+    def _prune_non_cond_memory(self, frame_idx, inference_state=None):
+        """Prune old non-conditioning frames to bound memory usage."""
+        if not self.clear_non_cond_mem:
+            return
+        inference_state = inference_state or self.inference_state
+
+        # Determine window size
+        min_frame = frame_idx - self.model.num_maskmem * self.model.memory_temporal_stride_for_eval
+        output_dict = inference_state["output_dict"]
+
+        # Prune global non_cond_frame_outputs
+        for f in [k for k in output_dict["non_cond_frame_outputs"] if k < min_frame]:
+            output_dict["non_cond_frame_outputs"].pop(f, None)
+
+        # Prune per-object non_cond_frame_outputs
+        for obj_output_dict in inference_state.get("output_dict_per_obj", {}).values():
+            for f in [k for k in obj_output_dict["non_cond_frame_outputs"] if k < min_frame]:
+                obj_output_dict["non_cond_frame_outputs"].pop(f, None)
+
 
 class SAM2DynamicInteractivePredictor(SAM2Predictor):
     """SAM2DynamicInteractivePredictor extends SAM2Predictor to support dynamic interactions with video frames or a
@@ -2423,6 +2444,7 @@ class SAM3VideoPredictor(SAM2VideoPredictor, SAM3Predictor):
                 inference_state=inference_state,
             )
             output_dict[storage_key][frame] = current_out
+            self._prune_non_cond_memory(frame, inference_state=inference_state)
         # Create slices of per-object outputs for subsequent interaction with each
         # individual object after tracking.
         self._add_output_per_object(frame, current_out, storage_key, inference_state=inference_state)
@@ -2475,14 +2497,12 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         hotstart_delay=0,
         hotstart_unmatch_thresh=3,
         hotstart_dup_thresh=3,
-        # Whether to suppress masks only within hotstart. If False, we can suppress masks even if they start before hotstart period.
-        suppress_unmatched_only_within_hotstart=True,
-        init_trk_keep_alive=0,
-        max_trk_keep_alive=8,
+        init_trk_keep_alive=30,
+        max_trk_keep_alive=30,
         min_trk_keep_alive=-4,
         # Threshold for suppressing overlapping objects based on recent occlusion
         suppress_overlapping_based_on_recent_occlusion_threshold=0.0,
-        decrease_trk_keep_alive_for_empty_masklets=False,
+        decrease_trk_keep_alive_for_empty_masklets=True,
         o2o_matching_masklets_enable=False,  # Enable hungarian matching to match existing masklets
         suppress_det_close_to_boundary=False,
         fill_hole_area=16,
@@ -2513,7 +2533,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         self.hotstart_delay = hotstart_delay
         self.hotstart_unmatch_thresh = hotstart_unmatch_thresh
         self.hotstart_dup_thresh = hotstart_dup_thresh
-        self.suppress_unmatched_only_within_hotstart = suppress_unmatched_only_within_hotstart
         self.init_trk_keep_alive = init_trk_keep_alive
         self.max_trk_keep_alive = max_trk_keep_alive
         self.min_trk_keep_alive = min_trk_keep_alive
@@ -2638,7 +2657,7 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
                 ) > 0
 
         # names = getattr(self.model, "names", [str(i) for i in range(pred_scores.shape[0])])
-        names = dict(enumerate(str(i) for i in range(pred_masks.shape[0])))
+        names = dict(enumerate(str(i) for i in range(pred_boxes.shape[0])))
         results = []
         for masks, boxes, orig_img, img_path in zip([pred_masks], [pred_boxes], orig_imgs, self.batch[0]):
             results.append(Results(orig_img, path=img_path, names=names, masks=masks, boxes=boxes))
@@ -2689,7 +2708,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         metadata = tracker_metadata_new["metadata"]
         removed_obj_ids = metadata["removed_obj_ids"]
         out["removed_obj_ids"] = removed_obj_ids
-        out["suppressed_obj_ids"] = metadata["suppressed_obj_ids"][frame_idx]
         out["frame_stats"] = frame_stats
         if self.masklet_confirmation_enable:
             status = metadata["masklet_confirmation"]["status"]
@@ -3597,7 +3615,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         overlap_pair_to_frame_inds = metadata["overlap_pair_to_frame_inds"]
         # removed_obj_ids: object IDs that are suppressed via hot-start
         removed_obj_ids = metadata["removed_obj_ids"]
-        suppressed_obj_ids = metadata["suppressed_obj_ids"][frame_idx]
 
         obj_ids_newly_removed = set()  # object IDs to be newly removed on this frame
         hotstart_diff = frame_idx - self.hotstart_delay if not reverse else frame_idx + self.hotstart_delay
@@ -3647,12 +3664,12 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
                     )
             if (
                 trk_keep_alive[obj_id] <= 0  # Object has not been matched for too long
-                and not self.suppress_unmatched_only_within_hotstart
                 and obj_id not in removed_obj_ids
                 and obj_id not in obj_ids_newly_removed
             ):
-                LOGGER.debug(f"Suppressing object {obj_id} at frame {frame_idx}, due to being unmatched")
-                suppressed_obj_ids.add(obj_id)
+                LOGGER.debug(f"Removing object {obj_id} at frame {frame_idx}, due to being unmatched")
+                # directly removed the object instead of suppressing it
+                obj_ids_newly_removed.add(obj_id)
 
         # Step 3: removed tracks that overlaps with another track for `hotstart_dup_thresh` frames
         # a) find overlaps tracks -- we consider overlap if they match to the same detection
@@ -3831,8 +3848,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
             "trk_keep_alive": defaultdict(int),  # This is used only for object suppression not for removal
             "overlap_pair_to_frame_inds": defaultdict(list),
             "removed_obj_ids": set(),
-            # frame_idx --> set of objects with suppressed outputs, but still continue to be tracked
-            "suppressed_obj_ids": defaultdict(set),
         }
         if self.masklet_confirmation_enable:
             # all the following are np.ndarray with the same shape as `obj_ids_all_gpu`

ultralytics/nn/autobackend.py

@@ -245,15 +245,20 @@ class AutoBackend(nn.Module):
             check_requirements(("onnx", "onnxruntime-gpu" if cuda else "onnxruntime"))
             import onnxruntime
 
-            providers = ["CPUExecutionProvider"]
-            if cuda:
-                if "CUDAExecutionProvider" in onnxruntime.get_available_providers():
-                    providers.insert(0, ("CUDAExecutionProvider", {"device_id": device.index}))
-                else:  # Only log warning if CUDA was requested but unavailable
-                    LOGGER.warning("Failed to start ONNX Runtime with CUDA. Using CPU...")
-                    device = torch.device("cpu")
-                    cuda = False
-            LOGGER.info(f"Using ONNX Runtime {onnxruntime.__version__} {providers[0]}")
+            # Select execution provider: CUDA > CoreML (mps) > CPU
+            available = onnxruntime.get_available_providers()
+            if cuda and "CUDAExecutionProvider" in available:
+                providers = [("CUDAExecutionProvider", {"device_id": device.index}), "CPUExecutionProvider"]
+            elif device.type == "mps" and "CoreMLExecutionProvider" in available:
+                providers = ["CoreMLExecutionProvider", "CPUExecutionProvider"]
+            else:
+                providers = ["CPUExecutionProvider"]
+                if cuda:
+                    LOGGER.warning("CUDA requested but CUDAExecutionProvider not available. Using CPU...")
+                    device, cuda = torch.device("cpu"), False
+            LOGGER.info(
+                f"Using ONNX Runtime {onnxruntime.__version__} with {providers[0] if isinstance(providers[0], str) else providers[0][0]}"
+            )
             if onnx:
                 session = onnxruntime.InferenceSession(w, providers=providers)
             else:
@@ -271,7 +276,10 @@ class AutoBackend(nn.Module):
             metadata = session.get_modelmeta().custom_metadata_map
             dynamic = isinstance(session.get_outputs()[0].shape[0], str)
             fp16 = "float16" in session.get_inputs()[0].type
-            if not dynamic:
+
+            # Setup IO binding for optimized inference (CUDA only, not supported for CoreML)
+            use_io_binding = not dynamic and cuda
+            if use_io_binding:
                 io = session.io_binding()
                 bindings = []
                 for output in session.get_outputs():
@@ -371,32 +379,33 @@ class AutoBackend(nn.Module):
             is_trt10 = not hasattr(model, "num_bindings")
             num = range(model.num_io_tensors) if is_trt10 else range(model.num_bindings)
             for i in num:
+                # Get tensor info using TRT10+ or legacy API
                 if is_trt10:
                     name = model.get_tensor_name(i)
                     dtype = trt.nptype(model.get_tensor_dtype(name))
                     is_input = model.get_tensor_mode(name) == trt.TensorIOMode.INPUT
-                    if is_input:
-                        if -1 in tuple(model.get_tensor_shape(name)):
-                            dynamic = True
-                            context.set_input_shape(name, tuple(model.get_tensor_profile_shape(name, 0)[2]))
-                        if dtype == np.float16:
-                            fp16 = True
-                    else:
-                        output_names.append(name)
-                    shape = tuple(context.get_tensor_shape(name))
-                else:  # TensorRT < 10.0
+                    shape = tuple(model.get_tensor_shape(name))
+                    profile_shape = tuple(model.get_tensor_profile_shape(name, 0)[2]) if is_input else None
+                else:
                     name = model.get_binding_name(i)
                     dtype = trt.nptype(model.get_binding_dtype(i))
                     is_input = model.binding_is_input(i)
-                    if model.binding_is_input(i):
-                        if -1 in tuple(model.get_binding_shape(i)):  # dynamic
-                            dynamic = True
-                            context.set_binding_shape(i, tuple(model.get_profile_shape(0, i)[1]))
-                        if dtype == np.float16:
-                            fp16 = True
-                    else:
-                        output_names.append(name)
-                    shape = tuple(context.get_binding_shape(i))
+                    shape = tuple(model.get_binding_shape(i))
+                    profile_shape = tuple(model.get_profile_shape(0, i)[1]) if is_input else None
+
+                # Process input/output tensors
+                if is_input:
+                    if -1 in shape:
+                        dynamic = True
+                        if is_trt10:
+                            context.set_input_shape(name, profile_shape)
+                        else:
+                            context.set_binding_shape(i, profile_shape)
+                    if dtype == np.float16:
+                        fp16 = True
+                else:
+                    output_names.append(name)
+                shape = tuple(context.get_tensor_shape(name)) if is_trt10 else tuple(context.get_binding_shape(i))
                 im = torch.from_numpy(np.empty(shape, dtype=dtype)).to(device)
                 bindings[name] = Binding(name, dtype, shape, im, int(im.data_ptr()))
             binding_addrs = OrderedDict((n, d.ptr) for n, d in bindings.items())
@@ -418,8 +427,7 @@ class AutoBackend(nn.Module):
             LOGGER.info(f"Loading {w} for TensorFlow SavedModel inference...")
             import tensorflow as tf
 
-            keras = False  # assume TF1 saved_model
-            model = tf.keras.models.load_model(w) if keras else tf.saved_model.load(w)
+            model = tf.saved_model.load(w)
             metadata = Path(w) / "metadata.yaml"
 
         # TF GraphDef
@@ -699,10 +707,7 @@ class AutoBackend(nn.Module):
 
         # ONNX Runtime
         elif self.onnx or self.imx:
-            if self.dynamic:
-                im = im.cpu().numpy()  # torch to numpy
-                y = self.session.run(self.output_names, {self.session.get_inputs()[0].name: im})
-            else:
+            if self.use_io_binding:
                 if not self.cuda:
                     im = im.cpu()
                 self.io.bind_input(
@@ -715,6 +720,9 @@ class AutoBackend(nn.Module):
                 )
                 self.session.run_with_iobinding(self.io)
                 y = self.bindings
+            else:
+                im = im.cpu().numpy()  # torch to numpy
+                y = self.session.run(self.output_names, {self.session.get_inputs()[0].name: im})
             if self.imx:
                 if self.task == "detect":
                     # boxes, conf, cls
@@ -839,7 +847,7 @@ class AutoBackend(nn.Module):
         else:
             im = im.cpu().numpy()
             if self.saved_model:  # SavedModel
-                y = self.model(im, training=False) if self.keras else self.model.serving_default(im)
+                y = self.model.serving_default(im)
                 if not isinstance(y, list):
                     y = [y]
             elif self.pb:  # GraphDef
@@ -884,8 +892,6 @@ class AutoBackend(nn.Module):
                     y[1] = np.transpose(y[1], (0, 3, 1, 2))  # should be y = (1, 116, 8400), (1, 32, 160, 160)
             y = [x if isinstance(x, np.ndarray) else x.numpy() for x in y]
 
-        # for x in y:
-        #     print(type(x), len(x)) if isinstance(x, (list, tuple)) else print(type(x), x.shape)  # debug shapes
         if isinstance(y, (list, tuple)):
             if len(self.names) == 999 and (self.task == "segment" or len(y) == 2):  # segments and names not defined
                 nc = y[0].shape[1] - y[1].shape[1] - 4  # y = (1, 32, 160, 160), (1, 116, 8400)

ultralytics/utils/__init__.py

@@ -65,6 +65,7 @@ RKNN_CHIPS = frozenset(
         "rv1103b",
         "rv1106b",
         "rk2118",
+        "rv1126b",
     }
 )  # Rockchip processors available for export
 HELP_MSG = """

ultralytics/utils/plotting.py

@@ -24,17 +24,6 @@ class Colors:
     This class provides methods to work with the Ultralytics color palette, including converting hex color codes to RGB
     values and accessing predefined color schemes for object detection and pose estimation.
 
-    Attributes:
-        palette (list[tuple]): List of RGB color tuples for general use.
-        n (int): The number of colors in the palette.
-        pose_palette (np.ndarray): A specific color palette array for pose estimation with dtype np.uint8.
-
-    Examples:
-        >>> from ultralytics.utils.plotting import Colors
-        >>> colors = Colors()
-        >>> colors(5, True)  # Returns BGR format: (221, 111, 255)
-        >>> colors(5, False)  # Returns RGB format: (255, 111, 221)
-
     ## Ultralytics Color Palette
 
     | Index | Color                                                             | HEX       | RGB               |
@@ -89,6 +78,17 @@ class Colors:
 
         For Ultralytics brand colors see [https://www.ultralytics.com/brand](https://www.ultralytics.com/brand).
         Please use the official Ultralytics colors for all marketing materials.
+
+    Attributes:
+        palette (list[tuple]): List of RGB color tuples for general use.
+        n (int): The number of colors in the palette.
+        pose_palette (np.ndarray): A specific color palette array for pose estimation with dtype np.uint8.
+
+    Examples:
+        >>> from ultralytics.utils.plotting import Colors
+        >>> colors = Colors()
+        >>> colors(5, True)  # Returns BGR format: (221, 111, 255)
+        >>> colors(5, False)  # Returns RGB format: (255, 111, 221)
     """
 
     def __init__(self):
@@ -203,6 +203,8 @@ class Annotator:
         if not input_is_pil:
             if im.shape[2] == 1:  # handle grayscale
                 im = cv2.cvtColor(im, cv2.COLOR_GRAY2BGR)
+            elif im.shape[2] == 2:  # handle 2-channel images
+                im = np.ascontiguousarray(np.dstack((im, np.zeros_like(im[..., :1]))))
             elif im.shape[2] > 3:  # multispectral
                 im = np.ascontiguousarray(im[..., :3])
         if self.pil:  # use PIL